Context Sherpa is a specialized platform for Context Engineering. It bridges the gap between developers and AI coding agents by providing LLMs with the precise, high-fidelity signals they need to operate with expert-level accuracy while dramatically reducing token consumption.
Agentic coding tools often struggle with either "hallucinating" codebase relationships or overwhelming their context windows with irrelevant files. Context Sherpa solves this through two primary goals:
- Increase Accuracy: Provide AI agents with strong, symbolic signals (definitions, references, and impact analysis) to ensure code changes are correct and idiomatic.
- Optimize Context: Using SCIP-based indexing and structural analysis, Context Sherpa allows agents to pinpoint exactly what they need, often reducing the tokens required for a task by up to 90% compared to raw file-searching.
Traditional tools like grep are fast but lack symbol awareness, while semantic search often returns noisy, over-processed results. Context Sherpa's tool suite is designed as a high-fidelity alternative:
- vs. Grep: SCIP indexing provides precise symbol-resolution (definitions vs. references), eliminating the guesswork of text-based search.
- vs. Semantic Search: Where vectors often fail to distinguish between similar-looking code, our Structural Analysis (ast-grep) uses the code's abstract syntax tree for exact pattern matching—no vector databases or text embeddings required.
- vs. Indexing: Unlike massive, centralized indexing services, Context Sherpa is local-first, lightweight, and requires zero cloud configuration or complex RAG infrastructure.
Of course the benefit of text embeddings in a vector database and other traditional methods is that they are more of a one size fits all solution. They can be used for many programming languages and even non-code files. Remember that a different SCIP tool must be used for each programming language. Lucky for you, you can use both. Context Sherpa will add SCIP tools for more languages in the future.
- Code Atlas Explorer: A premium GUI to visualize and inspect your codebase relationships.
- Agent Rule Management: Dynamically manage project-specific standards using
ast-grepand natural language feedback. - Integrated Local Reasoning: Connect to local providers (Ollama, LM Studio) to enable Tiered Inference. This architecture utilizes local models for high-frequency context distillation and semantic triage, drastically reducing the token noise and costs associated with sending raw source code to frontier models.
- Universal MCP Server: A high-performance "headless" mode that integrates directly with tools like Cursor, Cline, and Roo Code.
Context Sherpa is a single executable that adapts to your needs:
| Feature | GUI Mode | Headless (MCP) Mode |
|---|---|---|
| Code Atlas Explorer | ✅ Yes | ❌ No |
| Rule Visualizer/Editor | ✅ Yes | ❌ API only |
| Dependency Manager | ✅ Yes | ❌ Manual |
| MCP Tool Access | ✅ Indirectly | ✅ Directly via Agent |
| Resource Usage | Standard App | Ultra Lightweight |
Download the latest version for your platform from GitHub Releases.
- Windows: Download the
.exeand run it. - macOS: Download the app, move to
/Applications, and clear quarantine if necessary:xattr -d com.apple.quarantine /Applications/Context-Sherpa.app
- Linux: Download the binary and provide execution permissions (
chmod +x).
Important
Security Notice: Pre-built binaries are currently not signed. When running the application for the first time, you may encounter OS security warnings (e.g., Windows SmartScreen or macOS Gatekeeper). You will need to manually allow the application to run.
This is a known limitation of the current release process. Future pre-built binaries will undergo formal code signing for convenience. Building from source (instructions below) avoids these security warnings.
Context Sherpa is built with Go and Wails. For detailed building instructions, please see BUILDING.md.
- Build the Frontend:
cd frontend && npm install && npm run build && cd ..
- Build the GUI:
# Install Wails CLI if not already present go install github.com/wailsapp/wails/v2/cmd/wails@latest # Build the binary wails build
AI agents connected to Context Sherpa gain access to a powerful set of tools for structural, symbolic, and semantic analysis.
For a full list of available tools and their parameters, see TOOLS.md.
Add Context Sherpa to your AI agent's mcp_settings.json:
{
"mcpServers": {
"context-sherpa": {
"command": "context-sherpa",
"args": ["--projectRoot", "/path/to/your/project"]
}
}
}Context Sherpa is built on a three-layer "Tiered Inference" strategy that balances speed, cost, and intelligence:
- Tier 1: Deterministic Logic (SCIP / ast-grep): Instant, 0-token cost structural mapping. Precise symbolic analysis that provides the ground truth for your codebase.
- Tier 2: Integrated Local Reasoning (Ollama / LM Studio): local distillation and semantic triage on your own hardware. This layer offsets costs and drastically reduces the token burden on Tier 3 frontier models.
- Tier 3: Strategic Frontier Intelligence (Claude / Gemini): High-level architectural planning and complex refactoring directed by the "Big Brain" (your primary agent).
By offloading Tier 1 and Tier 2 tasks to Context Sherpa, you save up to 90% in token costs while significantly increasing the accuracy of Tier 3 strategic work.
Context Sherpa uses 3rd party tools to provide high-fidelity intelligence. These can be managed directly in the GUI Settings area:
- ast-grep: The core structural analysis engine.
- SCIP Indexers: Language-specific indexers for Go, TypeScript, and Python.
- Inference Engines: Connect to Ollama or LM Studio to enable semantic reasoning tools.
- GUI Guide - A visual tour of Context Sherpa's premium interface.
- MCP Tools Reference - Detailed documentation for all 20+ MCP tools.
- Building Context Sherpa - How to compile the GUI and Headless versions.
- Contributing Guide - How to help improve Context Sherpa.
Context Sherpa leverages several incredible open-source projects:
- ast-grep: The core engine for high-performance structural code analysis.
- Wails: The framework powering our cross-platform desktop experience.
- Sourcegraph SCIP: Providing the foundation for precise symbolic code intelligence via
scip-go,scip-typescript, andscip-python. - Local Inference: We are grateful to the communities behind Ollama and LM Studio for providing the infrastructure that powers our tiered semantic inference.
This project is licensed under the MIT License - see the LICENSE file for details.